Use Ifclite
Apps & Automationuse-ifcliteUse Ifclite is an Agent Skill for apps & automation published in the LTplus-AG/ifc-lite repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Work with IFC / BIM files (.ifc, .ifcx, IFC2X3 / IFC4 / IFC4X3 / IFC5) from the terminal using the ifc-lite CLI — inspect, query, validate (IDS), export (CSV/JSON/IFC/glTF/Parquet), create elements, merge, convert schema versions, diff, clash-check, run BCF collaboration, and script the SDK. Use whenever a task involves reading, analysing, generating, or editing IFC building models, or when the user mentions IFC, BIM, BCF, IDS, or buildingSMART. Also covers running ifc-lite as an MCP server for tool-calling agents.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/LTplus-AG/ifc-lite/blob/HEAD/.claude/skills/use-ifclite/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/use-ifclite, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Using ifc-lite
ifc-lite is a headless BIM toolkit for IFC files. It was built for LLM
terminals: every command takes --json, stdout is data, stderr is status,
and exit 0/1 means pass/fail. Prefer it over hand-parsing IFC text.
Start here (every session)
ifc-lite --help # all commands
ifc-lite schema # full SDK API as JSON (namespaces, methods, params) — read before writing eval/run code
ifc-lite info model.ifc --json # schema version, entity counts, storeys, top types
If ifc-lite isn't installed, run via npx @ifc-lite/cli <command> or install
with npm install -g @ifc-lite/cli.
Core workflows
# Query entities (use exact IFC type names; comma-separated for multiple)
ifc-lite query model.ifc --type IfcWall --json
ifc-lite query model.ifc --type IfcWall --where "Pset_WallCommon.IsExternal=true" --json
ifc-lite query model.ifc --type IfcWall --all --json # props, quantities, materials, classifications, relationships
ifc-lite query model.ifc --type IfcDoor --count
# Inspect one entity fully
ifc-lite props model.ifc --id 42
# Export
ifc-lite export model.ifc --format csv --type IfcWall --columns Name,GlobalId,Pset_WallCommon.FireRating --out walls.csv
ifc-lite export model.ifc --format ifc --schema IFC4 --out filtered.ifc
# Validate against IDS rules (exit 1 on failure)
ifc-lite ids model.ifc requirements.ids --json
# Create IFC from scratch (30+ element types)
ifc-lite create wall --start 0,0,0 --end 5,0,0 --height 3 --thickness 0.2 --out wall.ifc
# Combine / convert / compare
ifc-lite merge arch.ifc struct.ifc mep.ifc --out federated.ifc
ifc-lite convert model.ifc --schema IFC4 --out v4.ifc
ifc-lite diff v1.ifc v2.ifc --by-entity --json
# Structural sanity check
ifc-lite validate model.ifc --json
The power tools: eval and run
When no flag fits, drop into the full SDK. The bim object is the entire
@ifc-lite/sdk surface (discover it with ifc-lite schema).
ifc-lite eval model.ifc "bim.query().byType('IfcWall').count()"
ifc-lite eval model.ifc "bim.storeys().map(s => s.name)"
ifc-lite run analysis.js model.ifc # a .js file with `bim` available globally
ifc-lite ask model.ifc "how many walls?" maps plain-English questions to common
recipes — handy for quick one-offs, but eval is more precise.
For the complete command + flag reference (every command, all flags, the 30+
create element types, bsdd, clash, lod, view/analyze), read
cli-reference.md.
Rules that keep IFC output correct
These are non-negotiable when emitting IFC names or editing models:
- Exact IFC EXPRESS names, never aliases. Types like
IfcWallStandardCase, relationships likeIfcRelAggregates(notAggregates). Attributes in IFC PascalCase:GlobalId,Name,Description,ObjectType. - STEP type names are stored UPPERCASE. Don't hand-case them — query
results already carry the resolved IFC type in
EntityData.type(e.g.entity.type === 'IfcWall'), so read that instead of reconstructing it. - Single vs federated models are both first-class. When multiple files are
merged/federated, IDs are namespaced; resolve through the federation registry
rather than assuming
globalId === expressId. (Single-model fallback: they're equal.) --wherefilters usePsetName.PropName=Value(e.g.Pset_WallCommon.IsExternal=true).- Coordinates are IFC Z-up
[x, y, z]when creating elements.
Tips for agent use
- Always pass
--json; pipe tojqfor filtering. - Run
ifc-lite schemabefore writingeval/runcode. - Use
--countfor quick counts;--allto get full entity data in one call. - Use
create --from-json(reads stdin) for programmatic generation. - Don't loop calling per-entity extraction; fetch in bulk with
query --all.
Running as an MCP server (tool-calling agents)
For agents that speak the Model Context Protocol, @ifc-lite/mcp exposes the
same capabilities as ~70 typed tools (model/query/IDS/BCF/clash/export/viewer/
mutation) plus pre-baked prompts.
ifc-lite mcp ./model.ifc # stdio transport (default)
ifc-lite mcp ./model.ifc --read-only # hide all mutation tools
ifc-lite mcp ./arch.ifc ./struct.ifc --federate
ifc-lite mcp --transport http --port 8765 --token <secret>
# (equivalent standalone binary: ifc-lite-mcp …)
Tool families and the discovery-first calling pattern are documented in
mcp-tools.md.